Re: Constant WAL replay

[Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>]

=?ISO-8859-1?Q?Hans-J=FCrgen_Sch=F6nig?= <postgres(at)cybertec(dot)at> writes:
> We have toyed around with PostgreSQL's WAL structure and we are 
> wondering whether it is (theoretically) possible to replay WAL records 
> coming from a remote host while a system is working in read only mode.

There's been some idle speculation about that, and offhand I don't know
of any showstopper reasons why it couldn't be done ... but no one's
actually looked into making it happen.  Feel free ;-)

> Which kind of modifications are necessary to 
> replay and answer at the same time? (assuming that locks and 
> transactional information are available on the slave).

I think you should definitely assume that locks are *not* available, and
probably not any live transaction info.  The discussions before included
the idea of being able to do queries against a known-broken,
non-recoverable database without making it any worse than it was before.
In the current code there is no such thing as a hard read-only behavior
--- for example we will try to update commit-status hint bits no matter
what.  Allowing that to be turned off would be interesting for a number
of purposes, such as burning a database onto CD.

			regards, tom lane